When we started building Onbox, everyone asked:
Why not just add AI to Gmail? Why build a whole email client?
We chose the harder path. Here's why.
We can't stand broken experiences
If we're being honest, the real first reason is that we care too much about product to ship something broken.
At my last company, we tried automating email using Front (a popular support platform).
We built:
- a plugin to push emails and documents into our ERP
- a webhook to auto-tag buildings and co-owners
On paper, great. In practice:
- on small screens, users kept the plugin closed — Front already had too many columns
- even when open, it felt separate from where people actually worked
- one extra click was still one click too many, hurting adoption badly
Here's the real kicker: in Front, collaboration happens under the email, in comments between teammates. That's where people say "can you check this?" and "what should we do?".
With a plugin, our AI would sit next to that conversation. We want it in the conversation.
Also, for ops people, email is constant browsing, collaboration, and context switching. The baseline has to be:
- fast like Superhuman
- collaborative like Front (comments, shared context)
- with an AI you talk to in the thread, like a teammate you trust to do real work
The business case
Building on Gmail/Outlook sounds smart:
- ship faster
- live where people already are
But that math doesn't work for us:
But for our users, that doesn’t help as much as you’d think:
-
They don’t love Gmail/Outlook.
They use them because they’re the default, not because they’re good for their work. Many told us they’d happily switch for something that feels much better for their day-to-day. Some people love Front btw. But our experience showed that ops people are not on Front. -
We’d still need our own data layer.
Serious AI agents mean ingesting and indexing emails, attachments, and external tools on our side, organising them around workflows, and querying them a lot. Provider APIs and rate limits aren’t enough for that. -
So the only thing we’d really “save” is the client UI, with big UX constraints attached.
On the backend, we’d almost do the same work. The cost would be living forever inside someone else’s interface and limits.
We'd rather build something 10× better than something "good enough with AI on the side."
That's why Onbox is a full email client.
For the work we care about, the inbox has to be designed for AI + human working together — not just AI watching from the sidelines.